Introduction

John Oliver, in his very first episode on HBO's new show, Last Week Tonight, talked about Indian elections and the near-absence of its coverage in US Television. In response to that show, Alyssa Ayres wrote in Forbes that US's attention on India was scant. This, according to her, is also reflected in the pattern of languages studied in the US. She presented data in a tabular format from the Modern Languages Enrollment Database for 3 survey years-2002, 2006, and 2009 to show that the number of students enrolling to study Indian languages were very few in comparison to other popular languages.

Assuming that the knowledge of a language helps one understand culture in a better manner and could also potentially provide an advantage in international relations and business, it was of interest to me to see how different languages fared across a longer time frame. Here, I look at the pattern of language enrollments from MLA surveys of enrollments in US institutions of higher education using ggplot2 and rely on rCharts to facilitate comparison between any two or more languages across and within different years. This post has been generated using slidify and the code to reproduce it, along with the code and data for different charts can be found on github.

Data

Enrollment data for US higher education institutions were collected for 8 survey years spanning over two and half decades - 1983,1986,1990,1995,1998,2002,2006, and 2009. (One could go back in history further, but I chose to stop there for convenience.) Why only those years and not intermediate years? Those were the years when surveys were conducted and hence, data were provided only for those years.

Data were collected on the following languages - French, German, Italian, Japanese, Spanish, Arabic, Chinese, Korean, Portuguese, and Russian. Following Alyssa Ayres' procedure, data on multiple Indian languages were collected and then summed for a composite Indian language score for each year. These Indian languages include Bengali, Gujarati, Hindi, Hindi-Urdu, Kannada, Malayalam, Marathi, Punjabi, Tamil, Telugu, and Urdu. For the purposes of this post, I'll look at only the overall Indian language enrollment and not its individual components.

A look at the dataset in a tabular form

library(knitr)
lang = read.csv("languages.csv")
langmajor = lang[-c(2:12)]
kable(langmajor)

Year	Indian	French	German	Italian	Japanese	Spanish	Arabic	Chinese	Korean	Portuguese	Russian
2009	3929	215954	96270	80672	73328	862688	35083	60311	8511	11371	26814
2006	3034	206079	94181	78091	66076	821654	23956	51286	7145	10267	24810
2002	2323	201979	91100	63899	52238	745127	10584	34153	5211	8385	23921
1998	1506	199064	89013	49287	43141	656445	5505	28456	4479	6926	23791
1995	1232	205351	96263	43760	44723	606286	4444	26471	3343	6531	24729
1990	611	273116	133594	49824	45830	534143	3683	19427	2375	6118	44476
1986	522	275132	120920	40904	23454	411376	3417	16891	875	5071	33945
1983	408	270101	128140	38672	16127	386320	3441	13178	660	4397	30453

Pattern of Enrollments across the 8 years when surveys were conducted.

Before we see the patterns, we will need to convert our dataset into the long form. This, we perform using the reshape2 package. See how the dataframe format from the earlier table is modified to the following.

library(reshape2)
langmajormelt = melt(langmajor, id = "Year")
colnames(langmajormelt) = c("Year", "Language", "Enrollments")
kable(head(langmajormelt))

Year	Language	Enrollments
2009	Indian	3929
2006	Indian	3034
2002	Indian	2323
1998	Indian	1506
1995	Indian	1232
1990	Indian	611

kable(tail(langmajormelt))

id	Year	Language	Enrollments
83	2002	Russian	23921
84	1998	Russian	23791
85	1995	Russian	24729
86	1990	Russian	44476
87	1986	Russian	33945
88	1983	Russian	30453

On to the graph(s) --- ggplot2

library(ggplot2)
ggplot(langmajormelt, aes(x = Year, y = Enrollments)) + geom_point() + geom_line() + 
    facet_wrap(~Language, scales = "free") + theme_bw()

Interest in most languages has been on a rise. However, there were major drops in the interest in French, German, and Russian between 1990 and 1995.

Interactivity and Comparability Between Languages using rCharts

We could bring in interactivity to the charts by using the nvd3 JavaScript library via rCharts. In the chart below, we begin by plotting the interest in Indian languages across different survey years. You could interact with it many ways.

1. Clicking on a dot/circle of a legend can make that language appear/disappear from the plot. Can help with the comparison of enrollments between languages. you can have as many languages in the chart as you wish.
2. Hovering over the chart can provide information on the value during that year.

library(rCharts)
np1 = nPlot(Enrollments ~ Year, group = "Language", data = langmajormelt, type = "lineChart")
np1$set(disabled = c(F, T, T, T, T, T, T, T, T, T, T))
np1$xAxis(tickValues = c(1983, 1986, 1990, 1995, 1998, 2002, 2006, 2009))
np1$show("inline", include_assets = TRUE, standalone = TRUE)

Enrollments in each of the 8 years- Bar chart(s) from nvd3 via rCharts

In the barchart below, we begin by comparing enrollments in Chinese and in Indian languages across all years. You could interact with it in the following ways.

1. Clicking on a dot/circle of a legend can make that language appear/disappear from the plot. Can help with the comparison of enrollments between languages. you can have as many languages in the chart as you wish.
2. Hovering over the chart can provide information on the value for that bar and year.
3. You can make the comparison in either one specific year (select a year) or across all years (click on Year in the menu on the left)

np2 = nPlot(Enrollments ~ Year, group = "Language", data = langmajormelt, type = "multiBarChart")
np2$set(disabled = c(F, T, T, T, T, T, T, F, T, T, T))
np2$addFilters("Year")
np2$chart(showControls = F)
# np2$publish('np2',host='gist')

If this doesn't show up, please click on this link to interact with the chart